Privacy Policy

Records which apps you used and for how long — for example, “used VS Code for 2 hours” or “used Slack for 30 minutes.” This is the core function of Contextberg and forms the basis for all activity reports.

Records what you typed in which app. This enables analysis such as “spent 1 hour coding” or “wrote an email.” Collected input text is not used for any purpose other than productivity analysis. It is not sold to third parties or used for advertising.

Automatically takes snapshots of your screen every few minutes. Used to visually review your work and to help AI understand your work context. Captured images are stored on your PC and automatically deleted after 30 days.

Records text copied via copy operations. Used to help AI more accurately understand the work context during that time period. Can be disabled from the Settings screen.

Reads browsing history from Microsoft Edge and Google Chrome. Used to include web research and reference work in the activity log and to help AI understand the work context. Contextberg accesses the history files saved on your PC in read-only mode. It never modifies or writes to your browser settings or data. Can be disabled from the Settings screen.

Reads conversation logs stored on your PC by AI coding assistants such as Claude Code, Cursor, GitHub Copilot, and OpenCode. Used to help AI more accurately understand what you were thinking about and what implementation or research you were doing during that time period. Contextberg accesses these log files in read-only mode and never modifies or writes to the settings or conversation history of those tools. Can be disabled from the Settings screen.

After consent, Contextberg sends a randomly generated installation ID, app and build versions, CPU architecture, locale, and measurement environment (production, QA, or development) to AWS in the Tokyo region. The service uses receipt time to record the first-launch date and at most one daily-active record per day. The server hashes the installation ID before storage. This metadata does not include a device name, account information, screen content, input, clipboard content, browsing history, or work content. Daily event records expire automatically after no more than 400 days, and QA and development records are excluded from production usage metrics.

In Settings → Plugins, you choose Gmail or Google Calendar and authorize read-only access in Google. Gmail requests gmail.readonly, which grants mailbox-wide read access. You then choose one label and a 7, 14, or 30-day import window in ContextBerg. We read message and thread IDs, your account email, label names and IDs, subjects, senders, received dates, message body text and synchronization history IDs. Attachments, spam and trash are excluded from the cache. The label and period limit what the app imports; they do not narrow Google's mailbox-wide permission.

Calendar requests calendar.calendarlist.readonly so you can choose calendars, and calendar.events.readonly to read selected events. We read calendar IDs, names and time zones, and event IDs, titles, descriptions, locations, start/end times, update times and original links. Sync covers the previous 30 days and next 90 days. ContextBerg does not send, modify or delete Gmail messages, or create, change or delete Google events.

OAuth access and refresh tokens are stored in macOS Keychain. Calendar selections and events are stored in the local app database. The selected Gmail cache and Bot inbox, including source excerpts, conversations and reply threads, are encrypted locally with keys held in Keychain. Subject lines and event details are shown inside the app; macOS notification banners use generic text without private subjects, senders or event titles.

With Use in Bot enabled for a calendar, Bot can surface schedule changes and remind you before a timed meeting. Gmail notices use new messages in the chosen label. Refresh runs locally while the app is open; a sleeping or quit Mac delays detection. No remote push/webhook service receives your Google data for this feature.

Google authorization alone does not authorize sending Google content to AI. Bot asks you to confirm the selected AI destination for replies. Gmail also requires Allow Gmail context in Bot AI replies; changing your AI requires this consent again. Review updates with AI is optional and off by default. When permitted, selected email subjects, senders, body excerpts, event details and original source references may be included in replies or automatic review, alongside the conversation context. Separately enabled Calendar report/task options and Gmail task context may include selected source excerpts in derived reports or task suggestions.

The destination is the AI provider you select: Contextberg Cloud, through our AWS relay to Google Cloud Vertex AI; your configured Gemini or OpenRouter service; the provider associated with your Codex sign-in; or a local LM Studio model. A local model keeps inference on your Mac. External routes transmit the selected context to those services under their applicable terms. The Cloud relay does not persist prompt or response bodies; provider retention terms still apply. Restricted-scope review and any required security assessment for external processing must be completed before public rollout.

Deselecting a calendar removes its synced local events. Changing the Gmail label or import period replaces the prior cache. Disconnect removes that plugin's saved OAuth tokens and local source cache; it does not change your Google data or erase previously created Bot notices, threads, Tasks or reports. Hiding a Bot message is not deletion. Those separate records remain in local application storage until you remove the corresponding records or local app data. You can revoke Google access in your Google Account's third-party connections settings. For help removing data, contact support@contextberg.com.

Data sent for AI analysis passes through Contextberg’s relay server (AWS, Tokyo region) on its way to Vertex AI. Neither the relay server nor the Contextberg database stores the prompts (your work content) sent to Vertex AI or the responses (the generated report text) returned by Gemini. The transmitted and received data is discarded once the request completes, and your work content itself is never persisted in Contextberg’s storage.

However, the following operational metadata is recorded (it does not include the content itself): request timestamp, subscription plan, token counts, HTTP status codes, and error messages.

Your data is never used to train Gemini or any other AI model. Contextberg relies on Google Cloud Vertex AI under its Generative AI data governance, which guarantees that submitted prompts and responses are not used to train Vertex AI’s foundation models.

Google may retain prompts for abuse monitoring under its applicable data governance terms. See Google Cloud Vertex AI documentation for the applicable model and configuration.

As stated above, the Contextberg operator does not store your work content on the relay server or in the database, and therefore does not view your work content in normal operation.

However, within the scope of the technical logs and error traces that are available, the operator may temporarily access data for the following limited purposes only:

• Service quality improvement (incident investigation, bug reproduction, performance analysis)
• Identifying and investigating suspected violations of the terms of use or abusive usage

Such access is limited to the minimum necessary scope, and information obtained in this way is not used for any purpose other than those listed above.